GothamChess · Video summary

GPT-6 Astra vs Opus 5.5: The End Of Chess

The latest AI models, GPT-6 Astra and Claude Opus 5.5, face off in two chess games, revealing a surprising leap in their strategic capabilities and a few lingering quirks.

GothamChess · October 14, 2026 · 31-min episode · 3-min read

GPT-6 Astra vs Opus 5.5: The End Of Chess

Key takeaways

📈AI chess has dramatically improved, with models like GPT-6 Astra and Claude Opus 5.5 now playing at a level comparable to strong human players.00:31
📚Both AIs demonstrate a strong understanding of opening theory, consistently playing mainline moves for the first 10-12 moves of a game.01:54
🤔Despite their improvements, AIs can still miss complex tactical sequences, such as forced checkmates, opting for simpler, less decisive lines.11:14
🏆Claude Opus 5.5 showed impressive endgame conversion skills, systematically winning material and converting advantages into wins.13:39
🤝The AIs' ability to resign when in a clearly lost position indicates a sophisticated understanding of game state and material advantage.14:24
🧠The strategic depth of the games played suggests that these AIs are operating at a level above one-move blunders, engaging in higher-level strategic planning.18:37

Chapters

0:00Introduction to AI Chess Match
0:14Game 1: GPT-6 Astra (White) vs. Claude Opus 5.5 (Black)
0:15Game 2: Claude Opus 5.5 (White) vs. GPT-6 Astra (Black)
0:22Game 2: Opening and Strategic Plans
0:24Game 1: Opening and Early Middle Game
0:30Game 2: Illegal Move and Blunder
0:38Game 2: Opus's Dominance and Astra's Resignation
0:40Game 1: Critical Middle Game and Missed Tactics
0:45AI ELO Estimates and Final Thoughts
0:50Game 1: Endgame Conversion and Astra's Resignation

The digest

For years, the chess abilities of large language models like ChatGPT and Claude were a running joke. Prone to illegal moves, nonsensical piece placements, and a general lack of understanding of the game's fundamental rules, they were far from any serious competition. However, a recent two-game match between GPT-6 Astra and Claude Opus 5.5 reveals a dramatic shift: these AIs are now genuinely good at chess.

Game 1: Astra's D4 vs. Opus's Nimzo-Indian

The first game opened with GPT-6 Astra playing D4, a deliberate choice to avoid the common Sicilian Defense that AIs often default to after E4. Claude Opus responded with the Nimzo-Indian Defense, a sophisticated and theory-heavy opening. For the initial 10-12 moves, both AIs adhered strictly to known grandmaster theory, a stark contrast to their past blundering tendencies. This adherence to established lines demonstrated a significant improvement in their ability to access and utilize opening databases.

As the game progressed, Astra played a slightly inaccurate Queen E2, aiming to support a rook on D1. Opus, however, missed an opportunity to capture a hanging pawn, a decision that initially seemed like a mistake but ultimately led to a complex and balanced position. The game evolved into a rich middle game with an isolated queen's pawn for White, a common strategic theme in chess. Both AIs navigated this complexity without making any obvious blunders, showcasing a high level of positional understanding.